In reality, the smaller the study, the more likely teams are to rely on manual, spreadsheet-driven review to compensate for lean resourcing, which presents a huge risk.
Despite the best human efforts, manual review is a poor match for the way subject-level safety data now arrives: across multiple systems, in different formats, and often on different timelines. It leaves the process open to inconsistencies, and errors become easy to miss.
Here are the themes that kept resurfacing:
A small study does not produce small amounts of complexity. Even with fewer subjects, each patient can generate multiple data points including labs, vitals, concomitant medications, adverse events, protocol deviations, dosing changes, and follow-up notes, all of which must be interpreted together. If that data sits in separate trackers or exports, the reviewer has to rebuild the patient narrative from fragments, which is inefficient and error-prone.
This is of huge importance as the regulatory expectation under ICH E6(R3) doesn’t just expect safety data to exist , but to be assessed in a way that is proportionate, traceable, and fit for purpose. Small trial teams often underestimate how quickly fragmented review becomes unmanageable once the first dose-limiting concern, lab drift, or clustering symptom pattern appears.
Manual safety review asks reviewers to do several difficult things at once: spot change, remember what was reviewed last time, compare subjects, and decide whether a finding is meaningful. That is a heavy cognitive load to review and the process becomes especially fragile when the same reviewer has to rebuild context each time a new data cut arrives.
The hard reality is that the static nature of spreadsheets has no native awareness of what changed, what is new, or what was already reviewed. Once that context is lost, reviewers must spend their time re-checking old data instead of interpreting new risk, which is exactly the wrong allocation of effort for a safety process.
Safety issues in small trials rarely announce themselves loudly. More often, they show up as a slow drift in liver enzymes, a recurring symptom across visits, an accumulating pattern of dose delays, or a sequence of minor deviations that only becomes meaningful in longitudinal review. Manual reviewers struggles with this because human attention is naturally drawn to obvious outliers rather than subtle patterns spread across time.
FDA guidance on safety reporting further reinforces the need for timely review and escalation of relevant safety information, including sponsor assessment and investigator reporting pathways. If the review model only works when a concern is already obvious, it’s nothing more than a delayed confirmation.
Small trials often operate with limited bandwidth. In that environment, oversight is more vulnerable to inconsistency because different reviewers may interpret the same subject history slightly differently, especially when the data are incomplete or scattered. The result is not just delay, but variation in what gets escalated and when.
That variability is a real quality problem. ICH E6(R3) pushes the field toward risk-based quality management, proportionate oversight, and objective evidence that processes work as intended. A manual process that depends heavily on who is available and what they remember from the last cycle is difficult to defend as a robust system.
The hardest issue is that manual review can feel reassuring even when it is weak. A completed spreadsheet or signed checklist may suggest control, but doesn’t guarantee that the team saw the full subject trajectory or caught the most relevant change. In small trials, that false confidence is dangerous because there are fewer patients to absorb a mistake and less operational redundancy to catch it later.
This is the exact reason why risk-based oversight exists in the first place. It focusses attention on the data that matters most and creates a review process that is defensible, repeatable, and proportionate to study risk. Manual subject safety review breaks down when it tries to provide certainty without the tools to support it.
A better model is not more manual effort, but structured subject-level oversight that makes change visible, preserves context, and supports clear escalation when something moves outside expectation. For small trials, that usually means defining critical data upfront, standardizing review triggers, and building a workflow that highlights what changed since the last review rather than asking teams to re-read everything from scratch.
That is the real lesson here. That small trials are not low-risk by default, and safety review is not strong just because it is manual. The teams that manage subject safety well are the ones that make the review process fit the data, not the other way around.